[jira] [Created] (KAFKA-1062) Reading topic metadata from zookeeper leads to incompatible ordering of partition list

["Sam Meder (JIRA)" <ji...@apache.org>]

Sam Meder created KAFKA-1062:
--------------------------------

             Summary: Reading topic metadata from zookeeper leads to incompatible ordering of partition list
                 Key: KAFKA-1062
                 URL: https://issues.apache.org/jira/browse/KAFKA-1062
             Project: Kafka
          Issue Type: Bug
          Components: consumer
    Affects Versions: 0.8
            Reporter: Sam Meder
            Assignee: Neha Narkhede


The latest consumer changes to read data from Zookeeper during rebalance have made the consumer rebalance code incompatible with older versions (making rolling upgrades without downtime hard). The problem relates to how partitions are ordered. The old code seems to have returned the partitions sorted:

... rebalancing the following partitions: ArrayBuffer(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19) for topic produce-indexable-views with consumers: ...

the new code instead uses:

... rebalancing the following partitions: List(0, 5, 10, 14, 1, 6, 9, 13, 2, 17, 12, 7, 3, 18, 16, 11, 8, 19, 4, 15) for topic produce-indexable-views with consumers: ...

This causes new consumers and old consumers to claim the same partitions. I realize that this may not be a big deal (although painful for us since it disagrees with our deployment automation) since the code wasn't officially released, but it seems simple enough to sort the partitions if you'd take such a patch.


--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira